Tencent open‑sources Cube Sandbox

Tencent has open-sourced Cube Sandbox, a runtime for AI agents that the company says can start a hardware-isolated sandbox in under 60 milliseconds. (github.com) Cube Sandbox is built on RustVMM and Linux’s Kernel-based Virtual Machine, or KVM, which means each agent runs inside a lightweight virtual machine instead of a standard software container. Tencent says that keeps memory overhead below 5 megabytes per instance and lets the system scale from one machine to a cluster. (github.com) A sandbox is the locked room where an agent runs code, opens tools, or browses the web without touching the host machine. Tencent’s pitch is that Docker-style containers can be too exposed for model-written code, while full virtual machines are slower and heavier to launch. (news.ycombinator.com) Tencent says Cube is compatible with the E2B software development kit, so developers can switch by changing a single URL environment variable rather than rewriting application code. The GitHub repository also includes examples for the OpenAI Agents software development kit. (github.com) The timing lines up with a broader shift in agent infrastructure from chatbot demos to systems that spin up thousands of short-lived workers. Tencent Cloud’s Agent Runtime product page says its platform is aimed at coding agents, deep research, reinforcement learning, and graphical user interface agents, with “hundreds of thousands” of instances per minute and millisecond-level startup. (cloud.tencent.com) Tencent’s open-source portal lists Cube Sandbox as part of its public projects, and the GitHub repository shows the initial open-source release landed about five days ago, with documentation and example updates still arriving within the last several hours. That suggests Tencent is treating Cube as an active infrastructure project rather than a one-off code dump. (opensource.tencent.com) (github.com) Tencent engineers said on Hacker News that Cube had already been running inside Tencent Cloud before the public release, handling production AI agent workloads. They also said the team built it after finding closed-source sandbox services expensive and hard to self-host at large scale. (news.ycombinator.com) The bet is simple: if secure agent runtimes can start in tens of milliseconds instead of seconds or minutes, developers can afford to launch more agents only when needed. Cube Sandbox gives Tencent a public answer to that cost and latency problem, and now outside developers can test whether the numbers hold up. (github.com)